Page 140 - Artificial Intelligence in the Age of Neural Networks and Brain Computing
P. 140
4. Brain-Like AI Systems Based on SNN. NeuCube. 129
4.2.2 Semisupervised Learning
The proposed approach allows for training an SNN on a large part of data (unlabeled)
and training a classifier on a smaller part of the data (labeled), both datasets related to
the same problem. This is how the brain learns too.
Figs. 6.11 and 6.12 show the deep connectivity structures of a trained SNNcube on
fMRI and seismic data correspondingly (see Refs. [118,121,132,133]). Applications
for EEG data modelling are presented in [132] (see [6] for a review).
FIGURE 6.11
Deep learning of fMRI data in a NeuCube SNN. Voxels are mapped into a SNNcube using
Talairach template: (A) learned connections after STDP unsupervised learning using
affirmative sentence fMRI data represented by 20 selected voxels; (B) using negative
sentence data [121].
